CreatorFlowMethodology
Every number CreatorFlow shows is either measured, modelled or assumed. This page says which is which, and for each assumption, which direction it pushes the estimate. Nothing here is MRC-accredited measurement, and no figure on this platform is a promise of delivery.
What is measured
Audience retention comes from the creator's own connected YouTube Analytics, with their permission, through read-only access. YouTube reports retention as 100 points per video, so one point spans 1 percent of the video's length: 13.2 seconds on a 22-minute video, 54 seconds on a 90-minute one. We cannot resolve a slot position more finely than one point, and wherever a slot is placed the interface states the resolution it is working at.
We filter to organic views where it changes the result, and we read delivered segment impressions from YouTube's own reporting rather than recomputing them, so the delivery report reflects what YouTube measured, not what our model expected.
What is modelled
The view forecast for an upcoming video is fitted to the channel's trailing uploads of the same declared format, using their day-28 view counts. View counts are close to log-normal, so we fit that distribution and quote the 25th percentile, then take the cautious end of the confidence interval around it. With few uploads to fit, the quoted figure is visibly more conservative; below five comparable uploads we refuse to quote at all and the slot shows as unrated.
Prices are always quoted against that 25th percentile. The median never enters a price, and this is enforced in the codebase by tests, not by policy.
A note on the window: "trailing uploads at day 28" means the sample is at least a month old. For a weekly uploader the forecast looks back around ten weeks, which can straddle format changes and seasonal swings. The interface states the window it used.
What is assumed
On-target impressions multiply verified impressions by coverage: the share of the audience matching the buyer's target. YouTube provides age and gender together but geography separately, so coverage multiplies the two as if they were independent. They often are not. Where age profile and geography correlate, this overstates on-target reach, and that is the direction to correct in when reading the figure.
Audience composition is reported per channel, not per video. A collab or a breakout video can pull a different audience than the channel norm, and the estimate will not see it. Composition percentages also derive from signed-in viewing only.
For these reasons on-target impressions are labelled an estimate everywhere they appear. Delivery is verified against total verified impressions, which we can measure, not against the on-target estimate, which we cannot.
What is verified
When the sponsored video publishes, CreatorFlow detects it from the connected analytics, measures viewers present through the agreed slot window at day 14 and day 30, checks the paid-promotion disclosure, and writes the delivery report. Neither party enters the numbers and neither party can edit them. The report exists only for deals whose terms, including the declared price, were confirmed by both sides inside CreatorFlow before publication.
Flow Score
Flow Score is a 0 to 100 ranking aid for discovery, personalised to a buyer's target. It is guidance, it is never part of deal terms, and the codebase physically prevents it from reaching one. It weighs four components:
| Component | Weight | What it reads |
|---|---|---|
| Coverage | 35 | Share of the audience matching the buyer's target segment |
| Attention | 30 | Retention at the best mid-roll position against similar channels, less a penalty for costly ad reads |
| Consistency | 20 | How stable recent view counts are |
| Reliability | 15 | Delivery record on completed deals |
When a channel has fewer than two completed deals, Reliability is unrated and the remaining components are reweighted; we never fill the gap with an invented neutral value. Comparisons against similar channels are suppressed entirely when the comparison group has fewer than fifteen channels, and absolute figures are shown instead. Every Flow Score opens into its components and their raw inputs.
What this is not
CreatorFlow is not an accredited measurement vendor and does not claim to be. Forecasts are estimates sold at their cautious end. On-target impressions are planning figures with the biases stated above. Verified impressions on delivery reports are read from YouTube Analytics, which is itself the measurement authority here. Where our numbers disagree with YouTube's, YouTube's win.